Abstract: Tradional information retrieval models assume that keywords in ctueries are parallel, but the requirements of users should be abstracted to a series of keywords groups, and the sematic relations of keywords inside the group are closer than outside. This is "Term Proximity Information Retrieval" (TPIR) defined in this paper, and we presented a solution framework based on Weigh Matrix(WMSF). WMSF abstractes documents and ctueries to Weigh Matrix Representation of Document and Query Weigh Matrix, and then implements the TPIR based on the caculating of similarity between them. Empirical results show that WMSF is appropriate for TPIR compared with traditional information retrieval models which simplify the TPIR problems actually.

Key words: Information retrieval, Weigh matrix, Vector space model